A Noval Graph Convolutional Neural Network and Its Application in Power Load Forecasting

Abstract

Graph Convolutional Networks (GCNs) have recently gained significant attention in the field of graph neural networks due to their ability to learn node representations and handle graph-structured data. In this paper, we propose a novel GCN architecture that combines Inception modules and residual learning to enhance the expressive power and efficiency of the model. The proposed architecture is composed of multiple Inception modules with residual connections, which enable the network to learn hierarchical features from local and global graph neighborhoods. To further improve the performance of the model, we also incorporate skip connections that allow gradient flow across multiple layers. We evaluate our proposed model on benchmark datasets and demonstrate its superior performance compared to other models. Our results show that the proposed model achieves better accuracy and convergence speed while maintaining a low computational cost. The MAE of our model training results is 7.98$\%$ better than the GCN model and the RMSE is 21.6$\%$ better.